IoT and Artificial Intelligence for Fault Classification in High

Efficiency Motors

Carlos Guerrero

, Fernando Villegas

, William Oñate

and Gustavo Caiza

Universidad Politécnica Salesiana, Quito 170146, Ecuador

Keywords: Intelligent Classification, Incipient Failures, Standard Deviation Statistical Tool, Classification Deep Neural

Network, IoT.

Abstract: High-efficiency three-phase induction motors are used in most industrial production processes; however, its

malfunctioning may cause unexpected interruptions, putting at risk both manufacturing operations and

operators. Consequently, it is desired to diagnose in real-time the most common incipient failures that may

occur in this type of rotating machinery. Thus, this document presents a study of intelligent classification of

incipient failures in an induction motor, diagnosis that is visualized from a dashboard in the cloud through a

one-way IoT architecture. Using the traditional Park transform technique, torque (iq) and magnetizing (id)

currents were obtained and analysed through the standard deviation statistical tool, to identify the dispersion

of their operating amplitudes when the motor is at normal (H) or faulty (ECF and SC) operation conditions;

these values were normalized and provided as input data to a classification deep neural network. The results

given by this AI technique in the diagnosis, for both the iq and id components, showed a mean accuracy of

100% for SC and a mean classification error of 20% and 25% for H and ECF, respectively.

1 INTRODUCTION

Three-phase induction motors are very important in

industry since they take part of most production

processes, showing to be robust, low-cost and easy to

maintain (Bessous, 2020)(Torres et al., n.d.).

Nevertheless, their malfunctioning may cause

unexpected interruptions of operations (Otero et al.,

2020).

Some methods traditionally used for maintenance

of induction motors include: motor current and motor

voltage signature analysis (MCSA and MVSA),

Hilbert-Huang transform (HHT), continuous wavelet

transform (CWT) (Saucedo-Dorantes et al., 2017),

Park vector demodulation (PVD) (Oñate et al., 2022)

and statistical analysis (Torres et al., n.d.). However,

the operator should be trained to understand the

statistics of the plots, either time or frequency-

domain, considering that many of these publications

indicate possible ambiguities in the diagnosis. Being

aware of current manufacturing processes and the

https://orcid.org/0000-0002-8764-3847

https://orcid.org/0000-0002-0797-812X

https://orcid.org/0000-0001-6982-2502

https://orcid.org/0000-0002-8227-7227

parts that constitute the industry of this era (Castellino

et al., 2020), it is important to have intelligent systems

that operate in real-time and facilitate the decision-

making process for operators (Alberto et al., 2021).

Consequently, there are currently being developed

systems that integrate traditional and intelligent

methods, such as (Yang et al., 2016), (Prins et al.,

2018), (Ghosh et al., 2020) and (Zamudio-Ramirez et

al., 2020), with a predominance of MCSA and Park’s

vector modulus (PVM) for a single component; such

studies consider as focal point the diagnosis of broken

bars (BRB) and bearings (BF) failures, side-lining

other common incipient failures such as short-circuit

(SC) and eccentricity (EF), which account for 38%

and 10% of the cases, respectively (Bessous,

2020)(Otero et al., 2020)(Dhamal & Bhatkar, 2019).

Thus, the purpose of this work is to provide the

operator with access to a dashboard in the cloud for

visualizing failure diagnosis through a one-way

architecture. The system carries out the ADC of the

stator currents supplied to the motor, then extracts the

Guerrero, C., Villegas, F., OÃ

sate, W. and Caiza, G.

IoT and Artiﬁcial Intelligence for Fault Classiﬁcation in High Efﬁciency Motors.

DOI: 10.5220/0011949200003612

In Proceedings of the 3rd International Symposium on Automation, Information and Computing (ISAIC 2022), pages 405-409

ISBN: 978-989-758-622-4; ISSN: 2975-9463

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

405

iq and id components through Park transform, and

further identifies the operating amplitudes of the

motor currents with and without failure using the

standard deviation. This process will enable to feed

the input layer of a deep neural network for

classifying the operating state of the motor as being

in good or faulty conditions (H, ECF and SC).

This document is constituted by section 1

introduction, section 2 methodology and procedure,

section 3 analysis of results and finally conclusions in

section 4.

2 METHODOLOGY AND

PROCEDURE

Figure 1 shows the different stages that constitute the

failure diagnosis system, starting with the data

acquisition of the three-phase stator currents of the

motor, which is operated in a testbench as shown in

Fig. 1. Afterwards, Park transform is used to calculate

torque and magnetization currents that vary with

time, and the standard deviation is applied to this data

as a statistical tool to analyze its trend and enable the

implementation of a neural network for failure

diagnosis, which are further sent to a server in the

cloud for their visualization.

Figure 1: Block diagram of the proposed intelligent system

for detecting failures in a high-efficiency motor.

Figure 2 shows the test module, in which the

motor is coupled to a magnetic brake system

controlled in closed-loop to maintain the nominal

operation values.

Figure 2: Test module.

2.1 Data Acquisition

In the test module of Fig. 2, the data of the motor

currents are acquired and transformed into voltage

through a YHDC 3TA17-200 transducer. In order to

reduce the processing load on the Raspberry Pi boards

(Park transform, statistical tool and neural network),

an Arduino board was included to perform the ADC

with 10 bits and a sampling time of 10 ms. After

various experimental tests, a total of 60000 data were

acquired for each line of the three-phase system and

for each test (without failure, short-circuit and

eccentricity).

A WEG W22 high-efficiency motor is used in the

tests, intentionally causing failures such as: short-

circuit (SC) to a loop of a supply line, and eccentricity

(ECF) with a slope of 4.39° with respect to the shaft

of the magnetic brake. The motor features are

illustrated in Table 1.

Table 1: Features of the WEG W22 high-efficiency motor.

Features Valour

Number of phases 3

Number of poles 4

Nominal Voltage 220 / 380 - 440 v

Nominal Current 1.87 / 1.08 – 1.12 A

Power 0.37 KW – 0.5 HP

Frequency 60 Hz

Speed 1700 / 1725 RPM

ISAIC 2022 - International Symposium on Automation, Information and Computing

406

2.2 Park Transform and Statistical

Method

Once the data of the currents supplied to the motor

has been discretized, Park transform was applied to

obtain the torque (iq) and magnetization (id) currents

(Torres et al., n.d.)(Asad et al., 2018), which were

subject to a statistical method based on the standard

deviation to analyze the dispersion of the results when

the motor is in excellent and faulty operating

conditions (García, 2018).

2.3 Neural Network

Due to the large amount of data that may be acquired

during the operation of the motor, it was used a

classification model based on a deep neural network

with a supervised training technique. Such AI

technique consists of the following blocks: libraries

(NumPy, Keras, Pandas and TensorFlow), import the

iq and id components from an .xlsx file, split the data

for training (70%) and testing (30%), a network with

sequential architecture between input, hidden and

output layers, an SGD (Stochastic Gradient Descent)

training optimizer to stabilize the learning rate, and

thus establish experimentally a network with 3000

propagation correction cycles to reduce data losses.

Finally, the model with extension .h5 is implemented

in a board with Cortex processor. Figure 3 shows the

red squares that indicate the attributes that were

modified in the AI model for its execution.

Figure 3: AI model.

2.4 One-Way IoT Architecture

A communication architecture from the plant to a

dashboard was implemented to visualize the results of

the failure diagnosis in the Web; such architecture is

shown in Fig. 4. The tags corresponding to the results

of the neural network are sent through a Mosquitto

MQTT broker (Moreno Cerdà, 2018) to the Node-

Red service within the AWS platform (Chanthakit &

Rattanapoka, 2018), where the operator logs in

through authentication.

Figure 4: Communication architecture.

3 ANALYSIS AND RESULTS

Considering the MPU processing capacity, several

samples were used as input data to the sequential

model of the network, identifying that the component

that has the larger dispersion among the operating

amplitudes of the motor occurred for 50 samples, as

shown in Table 2., this is 70% (840 inputs) for

training and 30% (360 inputs) for experimental tests.

Table 2: Percentage operation amplitudes of the motor.

Number of

samples

W/O

Failure

Eccentricity

Short-

circuit

1000 100% 98.5% 90.78%

100 100% 98.49% 88.65%

50 100% 98.48% 87.98%

25 100% 98.74% 88.16%

10 100% 98.92% 88.27%

Number of

samples

W/O

Failure

Eccentricity

Short-

circuit

1000 100% 98.45% 90.66%

100 100% 98.44% 88.69%

50 100% 98.44% 88.08%

25 100% 98.63% 88.33%

10 100% 98.86% 88.52%

Figure 5 shows the percentages of correct answers

of the intelligent classification system during the

diagnosis of failures in a motor, after various

functional field tests evaluated at the torque (iq) and

magnetization (id) currents.

IoT and Artiﬁcial Intelligence for Fault Classiﬁcation in High Efﬁciency Motors

407

Figure 5: Result of motor operation.

Figure 5 shows some results to verify the

performance of the failure diagnosis using neural

networks. It was obtained a correct diagnosis in 80%

of the cases corresponding to the motor in good

condition. On the other hand, a correct diagnosis was

obtained in 75% of the cases corresponding to the

eccentricity failure, whereas a correct diagnosis was

achieved in 100% of the cases of the short-circuit

failure. Regarding the dispersion among the operating

amplitudes of the motor, Table 2 shows that the

dispersion of H with respect to ECF and SC was

1.52% and 12.02%, respectively.

4 CONCLUSIONS

In this work, a classification deep neural network was

used in conjunction with the standard deviation as a

statistical tool to define percentages of dispersion of

the operating amplitudes of the motor, obtaining a

difference of only 1.52% between H and ECF and of

12.02% between H and SC; these data were used in

the diagnosis, both for the iq and id components, with

a mean accuracy of 100% for SC and a mean

classification error of 20% and 25% for H and ECF,

respectively. The aforementioned results were

obtained with the experimental modification of

attributes in a deep neural classification model

constituted by 5 features in the input layer, each with

1200 input data (iq or id), a hidden layer with 1000

neurons and 5 outputs as classes corresponding to the

inputs. In order to contribute with the intelligent

system for diagnosing failures in induction motors, it

is foreseen to improve the amplitude of the operating

dispersions of the motor, and to avoid overlapping

conflicts in the system, it is possible to improve the

ADC of the data acquisition system.

REFERENCES

Alberto, J., De Almeida, A. T., & Ferreira, F. J. T. E.

(2021). Experimental study on the external shaft

axial stray flux in squirrel-cage induction motors.

Proceedings - 2021 IEEE Workshop on Electrical

Machines Design, Control and Diagnosis,

WEMDCD 2021, 254–259.

https://doi.org/10.1109/WEMDCD51469.2021.942

5627

Asad, B., Vaimann, T., Belahcen, A., & Kallaste, A. (2018).

Broken rotor bar fault diagnostic of inverter fed

induction motor using FFT, Hilbert and Park’s

Vector approach. Proceedings - 2018 23rd

International Conference on Electrical Machines,

ICEM 2018, 2352–2358.

https://doi.org/10.1109/ICELMACH.2018.8506957

Bessous, N. (2020). Reliability surveys of fault

distributions in rotating electrical machines : - Case

study of fault detections in IMS - Case s. CCSSP

2020 - 1st International Conference on

Communications, Control Systems and Signal

Processing, 535–543.

https://doi.org/10.1109/CCSSP49278.2020.9151672

Castellino, A., Donolo, P., de Angelo, C., & Bossio, G.

(2020). Análisis de las potencias instantáneas en

motores de inducción con excentricidad usando un

modelo con distribución sinusoidal. 2020 IEEE

Congreso Bienal de Argentina, ARGENCON 2020 -

2020 IEEE Biennial Congress of Argentina,

ARGENCON 2020.

https://doi.org/10.1109/ARGENCON49523.2020.9

505533

Chanthakit, S., & Rattanapoka, C. (2018). Mqtt based air

quality monitoring system using node MCU and

node-red. Proceeding of 2018 7th ICT International

Student Project Conference, ICT-ISPC 2018, 3–7.

https://doi.org/10.1109/ICT-ISPC.2018.8523891

Dhamal, S. S., & Bhatkar, M. V. (2019). Modelling and

simulation of three-phase induction motor to

diagnose the performance on inter-turn short circuit

fault in stator winding. 2018 International

Conference on Computing, Power and

Communication Technologies, GUCON 2018,

1166–1172.

https://doi.org/10.1109/GUCON.2018.8674900

García, J. (2018). Detección de fallas en motores trifásicos

de inducción utilizando análisis de componentes

independientes (ICA). 1–55.

Ghosh, A., Barman, P. K., & Das, S. (2020). Statistical

Feature Based Identification of Rotor Fault

Indicators for Three Phase Induction Motor.

Proceedings of 2020 IEEE-HYDCON International

Conference on Engineering in the 4th Industrial

Revolution, HYDCON 2020, 19–23.

https://doi.org/10.1109/HYDCON48903.2020.9242

691

Moreno Cerdà, F. (2018). Demostrador arquitectura

publish / subscribe con MQTT. 55.

Oñate, W., Gallardo, Y., Pérez, R., & Caiza, G. (2022).

ISAIC 2022 - International Symposium on Automation, Information and Computing

408

Comparative Analysis of High Frequencies for the

Broken Bar Fault Diagnosis Using MCSA and

Park’s Vector Demodulation. In Smart Innovation,

Systems and Technologies (Vol. 255). Springer

Singapore. https://doi.org/10.1007/978-981-16-

4884-7_10

Otero, M., Barrera, P., Bossio, G., & Leidhold, R. (2020).

Aplicacion de una estrategia activa de deteccion de

fallas en motores de induccion accionados por

variadores de velocidad comerciales. Congreso

Argentino de Control Automatico AADECA,

October, 6.

Prins, S., Mini, V. P., Mayadevi, N., & Harikumar, R.

(2018). Detection of Broken Rotor Bars Using

Multilevel Wavelet Decomposition. Proceedings of

the 2nd International Conference on Trends in

Electronics and Informatics, ICOEI 2018, Icoei,

621–626.

https://doi.org/10.1109/ICOEI.2018.8553821

Saucedo-Dorantes, J. J., Osornio-Rios, R. A., Delgado-

Prieto, M., & Romero-Troncoso, R. J. (2017).

Diagnosis methodology based on statistical-time

features and linear discriminant analysis applied to

induction motors. Proceedings of the 2017 IEEE

11th International Symposium on Diagnostics for

Electrical Machines, Power Electronics and Drives,

SDEMPED 2017, 2017-Janua(278033), 517–523.

https://doi.org/10.1109/DEMPED.2017.8062404

Torres, D., Oñate, W., & Caiza, G. (n.d.). Statistical

Analysis of Multidimensional Technique

Components for the Diagnosis of Incipient Failures

in Efficient High-Power Motors. 1–13.

Yang, T., Pen, H., Wang, Z., & Chang, C. S. (2016). Feature

Knowledge Based Fault Detection of Induction

Motors Through the Analysis of Stator Current Data.

IEEE Transactions on Instrumentation and

Measurement, 65(3), 549–558.

https://doi.org/10.1109/TIM.2015.2498978

Zamudio-Ramirez, I., Ramirez-Nunez, J. A., Antonino-

Daviu, J., Osornio-Rios, R. A., Quijano-Lopez, A.,

Razik, H., & De Jesus Romero-Troncoso, R. (2020).

Automatic Diagnosis of Electromechanical Faults in

Induction Motors Based on the Transient Analysis of

the Stray Flux via MUSIC Methods. IEEE

Transactions on Industry Applications, 56(4), 3604–

3613. https://doi.org/10.1109/TIA.2020.2988002

IoT and Artiﬁcial Intelligence for Fault Classiﬁcation in High Efﬁciency Motors

409